I am having trouble understanding how dataset caching works with clearml-data . I am using a clearml-task that runs python code to get a dataset from clearml-data . The worker is a Azure VM. I often run into issue where a dataset was downloaded on the machine, training ran for a few hours, but then dataset is no longer on the machine for validation (and the script crash and model artifact is never uploaded). How / when is dataset cache cleared ?